A Comprehensive Survey on Deep Learning-Based LiDAR Super-Resolution for Autonomous Driving
New survey maps deep learning techniques to make cheap LiDAR sensors see like expensive ones.
Researchers June Moh Goo, Zichao Zeng, and Jan Boehm have published the first comprehensive survey on deep learning-based LiDAR super-resolution for autonomous driving. The paper, accepted to IEEE IV 2026, systematically reviews four key method categories—from CNNs to Transformer and Mamba-based approaches—and establishes benchmarks. It highlights a major industry trend toward real-time inference and cross-sensor generalization, enabling cheaper, low-resolution LiDAR sensors to produce detailed, high-resolution 3D point clouds for self-driving cars.
Why It Matters
This technology could drastically reduce the cost of autonomous vehicle perception systems, accelerating widespread deployment.